Defining mediation effects for multiple mediators using the concept of the target randomized trial
share
Causal mediation approaches have been primarily developed for the goal of "explanation", that is, to understand the pathways that lead from a cause to its effect. A related goal is to evaluate the impact of interventions on mediators, for example in epidemiological studies seeking to inform policies to improve outcomes for sick or disadvantaged populations by targeting intermediate processes. While there has been some methodological work on evaluating mediator interventions, no proposal explicitly defines the target estimands in terms of a "target trial": the hypothetical randomized controlled trial that one might seek to emulate. In this paper, we define so-called interventional effects in terms of a target trial evaluating a number of population-level mediator interventions in the context of multiple interdependent mediators and real-world constraints of policy implementation such as limited resources, with extension to the evaluation of sequential interventions. We describe the assumptions required to identify these novel effects from observational data and a g-computation estimation method. This work was motivated by an investigation into alternative strategies for improving the psychosocial outcomes of adolescent self-harmers, based on data from the Victorian Adolescent Health Cohort Study. We use this example to show how our approach can be used to inform the prioritization of alternative courses of action. Our proposal opens up avenues for the definition and estimation of mediation effects that are policy-relevant, providing a valuable tool for building an evidence base on which to justify future time and financial investments in the development and evaluation of interventions.
READ FULL TEXT